[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"cheat-sheet---en":3,"domain-info---en":3,"topic-info----en":3,"lesson-aws-certified-ai-practitioner-genai-fundamentals-genai-business-value-what-genai-does-well-en":4,"next-aws-certified-ai-practitioner-genai-fundamentals-genai-business-value-what-genai-does-well-en":364,"prev-aws-certified-ai-practitioner-genai-fundamentals-genai-business-value-what-genai-does-well-en":376},null,{"locked":5,"reason":3,"meta":6,"item":17},false,{"title":7,"description":8,"isFree":9,"estimatedMinutes":10,"difficulty":11,"learningObjectives":12},"What Generative AI Does Well","The four advantages the exam names, grounded in what a foundation model can do that a purpose-trained model cannot: adapt without retraining, respond without a build cycle, hold a conversation, and produce content that did not exist.",true,16,"beginner",[13,14,15,16],"Describe the advantages of generative AI named in the exam guide: adaptability, responsiveness, conversational capabilities, and the ability to generate content","Explain why a pre-trained foundation model removes the labeled-data and retraining cost that traditional ML carries","Compare a foundation model against a purpose-trained model on a task both could perform","Separate the advantages the exam attributes to the technology from those it attributes to the AWS platform",{"id":18,"title":7,"body":19,"description":8,"difficulty":11,"estimatedMinutes":10,"extension":301,"infographics":302,"isFree":9,"learningObjectives":303,"meta":304,"navigation":9,"path":305,"quiz":306,"seo":361,"stem":362,"__hash__":363},"courses/courses/aws-certified-ai-practitioner/en/domains/02-genai-fundamentals/02-genai-business-value/01-what-genai-does-well.md",{"type":20,"value":21,"toc":289},"minimark",[22,26,29,32,37,40,43,46,49,57,61,64,67,70,77,81,84,87,90,94,97,100,103,107,110,113,201,204,208,211,239,242,262,266,286],[23,24,25],"p",{},"A team spends six weeks building a sentiment classifier for product reviews. They label 20,000 reviews, train a model, tune it, deploy it. It works. Then the product manager asks for the same sentiment read on support emails, plus a topic tag on every review, plus a draft reply for the negative ones.",[23,27,28],{},"With the classifier, those are three new projects, each with its own labeling effort. With a foundation model, they are three sentences added to a prompt, working that afternoon.",[23,30,31],{},"That gap is the first of four advantages the exam guide names in Task Statement 2.2: \"adaptability, responsiveness, conversational capabilities, ability to generate content.\" Each one reads like marketing until you price it against what the previous approach cost.",[33,34,36],"h2",{"id":35},"adaptability-the-task-moves-out-of-training","Adaptability: the task moves out of training",[23,38,39],{},"A supervised model's task is decided by its labels, and labels are chosen before training starts. Train on positive and negative review labels and the model returns one of two classes for a review. Forever. Ask it for a topic tag and there is no mechanism for it to answer, because nothing in its output layer can represent a topic.",[23,41,42],{},"A foundation model works the other way round. Pre-training gave it broad capability with no particular task attached, so the task arrives at inference time, written in the prompt. Same weights, new instruction, different job.",[23,44,45],{},"The practical consequence is a change in the unit of work. In traditional ML, changing what the system does means a training run: gather labels, train, evaluate, redeploy, measured in days or weeks. With a foundation model it means editing text, measured in minutes. Teams feel this as a shift in what is worth trying, because a task that would never justify a six-week project is easy to justify as a paragraph.",[23,47,48],{},"Two boundaries keep this honest. Adaptability is not accuracy: a classifier trained on 20,000 labeled reviews will often beat a general model on those reviews specifically, and that is a real trade. And adaptability is not free: the prompt that carries the instruction is billed on every request, forever, while the classifier's training cost was paid once.",[23,50,51,52,56],{},"The exam has a name for the measurement of this property, ",[53,54,55],"strong",{},"cross-domain performance",", and it appears in the business-metrics objective covered in the last lesson of this topic.",[33,58,60],{"id":59},"responsiveness-two-meanings-both-tested","Responsiveness: two meanings, both tested",[23,62,63],{},"\"Responsiveness\" is the advantage most likely to be misread, because two different properties travel under the name.",[23,65,66],{},"The first is latency. A foundation model answers an arbitrary request in real time, typically streaming tokens as it produces them, so a person can read the beginning of an answer while the end is still being generated. That is what makes live conversation possible at all. AWS also offers latency-optimized inference on Amazon Bedrock for selected models, which routes requests to a faster serving path with no change to the model or the prompt.",[23,68,69],{},"The second is responsiveness to change: the system answers a request nobody anticipated, without a build cycle sitting between the request and the answer. When a new product launches on Tuesday, the assistant can discuss it on Tuesday, as long as the details reach the prompt.",[23,71,72,73,76],{},"Here is the misconception worth naming, because it is a tempting distractor. Responsiveness does ",[53,74,75],{},"not"," mean a foundation model is faster than the alternative. A small purpose-trained classifier returns in milliseconds; a large language model takes hundreds of milliseconds to several seconds and gets slower as the answer gets longer. If a scenario demands a decision in under 50 milliseconds inside a payment flow, generative AI is the wrong tool and the exam expects you to say so. Responsiveness means fast enough to hold a conversation and fast to change, not fastest in absolute terms.",[33,78,80],{"id":79},"conversational-capability-natural-language-on-both-ends","Conversational capability: natural language on both ends",[23,82,83],{},"The third advantage is the one users notice, and the reason generative AI reached non-technical people so quickly. The input is a sentence rather than a form, a schema, or a query language. The output is a sentence rather than a class label or a number.",[23,85,86],{},"Multi-turn is where it earns its place. An analyst asks \"which contracts renew this quarter?\", then \"just the ones over $100k\", then \"sort those by renewal date and tell me which have auto-renew clauses.\" Each follow-up is meaningless on its own. Together they are a conversation, and the model resolves \"those\" and \"the ones\" against what came before.",[23,88,89],{},"Now the mechanism, because the exam likes the gap between how this feels and how it works. The model has no memory. Frozen weights, one request at a time. What creates the illusion is that your application resends the earlier turns as input tokens with every new request, so the history arrives as part of the prompt. Three consequences follow directly, and all three are exam material: long conversations cost progressively more, a conversation eventually exceeds the context window, and nothing is remembered after the session ends unless you deliberately store it.",[33,91,93],{"id":92},"generating-content-output-that-did-not-exist-before","Generating content: output that did not exist before",[23,95,96],{},"The fourth advantage is the one that names the whole category. A traditional model selects: it picks a class, scores a probability, predicts a number that already had a right answer in the data. A generative model composes something new.",[23,98,99],{},"That covers the obvious outputs, text, images, audio, video, and code, and one that is easy to miss. AWS lists generating synthetic labeled data as a business process application, which means one useful job for generative AI is manufacturing training data for traditional machine learning. A team short on labeled fraud examples can generate plausible ones to strengthen a classifier. That inversion shows up in scenario questions.",[23,101,102],{},"Generation is also where the risks concentrate, for exactly the reason it is powerful: nothing constrains the output against a source. The next lesson is about that.",[33,104,106],{"id":105},"the-advantage-underneath-the-four-nothing-to-label","The advantage underneath the four: nothing to label",[23,108,109],{},"Pull the four advantages apart and the same fact sits underneath most of them. The expensive part already happened, in someone else's data center, months ago.",[23,111,112],{},"Put the two approaches side by side on the same task, sentiment on product reviews:",[114,115,116,131],"table",{},[117,118,119],"thead",{},[120,121,122,125,128],"tr",{},[123,124],"th",{},[123,126,127],{},"Purpose-trained classifier",[123,129,130],{},"Foundation model",[132,133,134,146,157,168,179,190],"tbody",{},[120,135,136,140,143],{},[137,138,139],"td",{},"Data needed before the first result",[137,141,142],{},"About 20,000 labeled reviews",[137,144,145],{},"None",[120,147,148,151,154],{},[137,149,150],{},"Time to a working result",[137,152,153],{},"Weeks",[137,155,156],{},"Hours",[120,158,159,162,165],{},[137,160,161],{},"Adding a second task",[137,163,164],{},"New labels, new training run, new deployment",[137,166,167],{},"One more sentence in the prompt",[120,169,170,173,176],{},[137,171,172],{},"Cost shape",[137,174,175],{},"Training paid once, inference very cheap",[137,177,178],{},"No training, per-token inference on every request",[120,180,181,184,187],{},[137,182,183],{},"Accuracy on that one narrow task",[137,185,186],{},"Usually higher",[137,188,189],{},"Usually good enough, sometimes better",[120,191,192,195,198],{},[137,193,194],{},"Who can change its behavior",[137,196,197],{},"An ML engineer",[137,199,200],{},"Anyone who can write clear instructions",[23,202,203],{},"The row that decides real projects is the cost shape. A foundation model removes the upfront investment and replaces it with a per-request charge that never ends. At 500 requests a day that trade is obviously good. At 50 million requests a day the classifier's cheap inference can win by a wide margin, even counting the labeling effort. Volume is what flips the answer, and a scenario that mentions very high request volume with a narrow, stable task is usually steering you away from a foundation model.",[33,205,207],{"id":206},"reading-the-signals-in-a-scenario","Reading the signals in a scenario",[23,209,210],{},"Four signals push toward generative AI, and they are worth carrying as a set:",[212,213,214,221,227,233],"ul",{},[215,216,217,220],"li",{},[53,218,219],{},"Unstructured input"," that no schema fits: free text, images, audio, mixed documents.",[215,222,223,226],{},[53,224,225],{},"Variable output"," where the answer is a paragraph or an image, with no single correct response to train against.",[215,228,229,232],{},[53,230,231],{},"Several related tasks"," that would each need their own model.",[215,234,235,238],{},[53,236,237],{},"No labeled dataset",", and no cheap way to build one.",[23,240,241],{},"Three push away:",[212,243,244,250,256],{},[215,245,246,249],{},[53,247,248],{},"One narrow prediction at very high volume"," with plenty of labels, where a small model is cheaper and more accurate.",[215,251,252,255],{},[53,253,254],{},"A hard latency budget"," measured in tens of milliseconds.",[215,257,258,261],{},[53,259,260],{},"A requirement that the same input always produce the same output",", which is the subject of the next lesson.",[33,263,265],{"id":264},"exam-tips","Exam tips",[212,267,268,271,274,277,280,283],{},[215,269,270],{},"Learn the four named advantages as written: adaptability, responsiveness, conversational capabilities, ability to generate content. Distractor lists usually swap in accuracy, determinism, or interpretability, which are the exact properties generative AI is weakest on.",[215,272,273],{},"Keep Task Statement 2.2 and 2.3 apart. Advantages of generative AI as a technology means the four above. Advantages of AWS generative AI services means accessibility, lower barrier to entry, efficiency, cost-effectiveness, and speed to market. The question wording tells you which list it wants.",[215,275,276],{},"Adaptability is about specifying the task in the prompt instead of in the labels. Any option describing retraining is describing the traditional approach.",[215,278,279],{},"Responsiveness does not mean lower latency than every alternative. A scenario with a strict millisecond budget on a single narrow prediction points at traditional ML.",[215,281,282],{},"Conversational memory is resent context, not stored state. Questions about rising cost in long conversations, or about hitting the context window, come from this fact.",[215,284,285],{},"Watch for very high volume plus a narrow stable task. That combination favors a purpose-trained model even though a foundation model could do the job.",[23,287,288],{},"One idea to carry: every advantage in this lesson comes from the model being general and already trained, and every limitation in the next lesson comes from the same two facts. Generality is why it adapts, and it is also why it has no idea when it is wrong.",{"title":290,"searchDepth":291,"depth":291,"links":292},"",3,[293,295,296,297,298,299,300],{"id":35,"depth":294,"text":36},2,{"id":59,"depth":294,"text":60},{"id":79,"depth":294,"text":80},{"id":92,"depth":294,"text":93},{"id":105,"depth":294,"text":106},{"id":206,"depth":294,"text":207},{"id":264,"depth":294,"text":265},"md",[],[13,14,15,16],{},"/courses/aws-certified-ai-practitioner/en/domains/02-genai-fundamentals/02-genai-business-value/01-what-genai-does-well",{"passingScore":307,"questions":308},70,[309,318,326,334,342,353],{"question":310,"type":311,"options":312,"correctAnswer":314,"explanation":317},"Which four advantages of generative AI does Task Statement 2.2 of the exam guide name?","single",[313,314,315,316],"Accuracy, determinism, interpretability, and low latency","Adaptability, responsiveness, conversational capabilities, and the ability to generate content","Security, compliance, redundancy, and regional coverage","Speed to market, lower barrier to entry, accessibility, and cost-effectiveness","The exam guide lists exactly those four as advantages of generative AI. Accuracy and determinism are not advantages of generative AI at all, and the fourth option is the separate list of advantages of using AWS generative AI services from Task Statement 2.3, which the exam keeps distinct.",{"question":319,"type":311,"options":320,"correctAnswer":322,"explanation":325},"A team has a sentiment classifier trained on 20,000 labeled product reviews. They now also need a topic tag for each review. What does adaptability mean for this situation?",[321,322,323,324],"The classifier can be retrained on the same labels to produce topic tags","A foundation model can produce topic tags from an added instruction in the prompt, with no new labeling or training run","The classifier will automatically start producing topic tags once it sees enough reviews","A foundation model will produce more accurate sentiment than the classifier","Adaptability means the task is specified at inference time rather than fixed at training time, so a new task costs a prompt edit instead of a project. The classifier is locked to the label set it was trained on, and adaptability makes no promise about beating a purpose-trained model on the one narrow task it was built for.",{"question":327,"type":311,"options":328,"correctAnswer":331,"explanation":333},"Why does a foundation model usually produce a working result faster than a traditional supervised model for a new text task?",[329,330,331,332],"Because foundation models run on faster hardware","Because foundation models do not make mistakes","Because pre-training already happened, so the project starts with a capable model and no labeled dataset","Because foundation models are smaller than classifiers","The weeks a traditional project spends collecting labels and training are already paid for in the model's pre-training, so the first useful output can arrive the same day. Foundation models are far larger than most classifiers, generally slower per request, and certainly not error-free.",{"question":335,"type":311,"options":336,"correctAnswer":338,"explanation":341},"A support assistant remembers what a customer said three messages ago. How is that memory actually implemented?",[337,338,339,340],"The model stores the conversation in its weights as it goes","The earlier turns are resent as input tokens with each new request","The model queries a database of past conversations automatically","The model has permanent memory of every user it has spoken with","Conversational capability comes from carrying the history forward in the prompt, which is why long conversations cost more and eventually hit the context window. Model weights are frozen at inference time, and nothing is remembered between sessions unless the application deliberately stores and resends it.",{"question":343,"type":344,"options":345,"correctAnswers":351,"explanation":352},"Which of these are genuine advantages of using a foundation model rather than a purpose-trained model? (Choose 2.)","multiple",[346,347,348,349,350],"It reliably achieves higher accuracy than a specialist model on that specialist's single task","One model handles several related tasks that would otherwise need several models","It produces the same output every time for the same input","It can start producing useful output without a labeled training dataset","It costs less per request at very high volume",[347,349],"Cross-task coverage and the removal of the labeling requirement are the two advantages that hold up consistently. A specialist model trained on plenty of labeled data often wins on its own narrow task, foundation models are nondeterministic by default, and per-request inference cost is usually the place where a small purpose-trained model wins at scale.",{"question":354,"type":311,"options":355,"correctAnswer":356,"explanation":360},"An exam question asks about the advantages of building on Amazon Bedrock rather than self-hosting. Which list should you answer from?",[356,357,358,359],"The Task Statement 2.3 list: accessibility, lower barrier to entry, efficiency, cost-effectiveness, speed to market","The Task Statement 2.2 list: adaptability, responsiveness, conversational capabilities, content generation","The responsible AI dimensions","The foundation model lifecycle stages","The exam separates advantages of the technology from advantages of the AWS services that deliver it, and the question wording tells you which one is being asked. Reading a platform question and answering with adaptability and conversational capability is a common way to lose a mark you understood perfectly well.",{"title":7,"description":8},"courses/aws-certified-ai-practitioner/en/domains/02-genai-fundamentals/02-genai-business-value/01-what-genai-does-well","nONGnhW95C8qFiI791F5Ul-x7EMn_6ns1CcAsgLu4Qc",{"locked":9,"reason":365,"meta":366,"item":3},"paywall",{"title":367,"description":368,"isFree":5,"estimatedMinutes":369,"difficulty":11,"learningObjectives":370},"Limitations and Risks of Generative AI","Hallucination, inaccuracy, nondeterminism, and interpretability are four different failures with four different fixes. Learn to tell them apart, and see how Amazon Bedrock detects the one you cannot prevent.",18,[371,372,373,374,375],"Identify the disadvantages of generative AI named in the exam guide: hallucinations, interpretability, inaccuracy, and nondeterminism","Distinguish a hallucination from a plain inaccuracy, and both from output variation","Explain how temperature, top-p, and top-k produce nondeterministic output and what lowering them does","Describe how the Amazon Bedrock Guardrails contextual grounding check scores grounding and relevance","Match each failure mode to the mitigation that actually addresses it",{"locked":9,"reason":365,"meta":377,"item":3},{"title":378,"description":379,"isFree":5,"estimatedMinutes":380,"difficulty":381,"learningObjectives":382},"Agentic AI and the Model Context Protocol","See what turns a foundation model into an agent: the cyclical reasoning loop, tool use, memory, and multi-agent patterns, plus how the Model Context Protocol standardizes the connection between agents and the systems they act on.",22,"intermediate",[383,384,385,386],"Define an AI agent and explain what separates it from a single foundation model call","Trace an agent's reasoning loop through tool selection, action, and observation","Describe memory management and common multi-agent patterns","Explain what the Model Context Protocol standardizes and how AWS supports it"]